Using convolutional neural network to analyze brain MRI images for predicting functional outcomes of stroke

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Physicians currently rely on clinical experience and big data for stroke outcome prediction.
• Accurate prediction of functional outcomes is crucial for personalized stroke patient management.
• Identifying specific imaging features can improve the precision of outcome prognostication.

Purpose of the Study:

• To develop and validate a deep learning model for predicting functional outcomes in stroke patients.
• To identify key magnetic resonance imaging (MRI) features associated with post-stroke recovery.
• To create an automated decision support system for clinical use in stroke care.

Main Methods:

• Utilized magnetic resonance imaging (MRI) data from ischemic and hemorrhagic stroke patients.
• Developed and trained a VGG-16 convolutional neural network (CNN) model.
• Assessed functional outcomes using the modified Rankin Scale (mRS) and National Institutes of Health Stroke Scale (NIHSS).
• Trained separate models for men, women, and mixed groups to evaluate performance differences.

Main Results:

• The VGG-16 CNN model demonstrated high accuracy in predicting functional outcomes after 28-day hospitalization.
• The deep learning approach successfully identified imaging features relevant to stroke prognosis.
• Analysis showed comparable performance across gender-specific and mixed-gender training groups.

Conclusions:

• A VGG-16 CNN model effectively predicts stroke patient functional outcomes using MRI data.
• This deep learning approach provides a valuable automated decision support tool for clinicians.
• The system facilitates personalized recommendations and treatments, enhancing clinical practice for stroke patients.